A robot system watches its surroundings, predicts what may happen next, and changes its actions in real time to handle moving objects and changing work conditions.
Robotics startup Rhoda AI has revealed a robotics system designed to help robots work in changing real-world environments rather than controlled lab settings. The system, called FutureVision, predicts how the physical world will change and converts those predictions into robot actions. It observes the environment, generates short video predictions of what will happen next, acts based on those predictions, and repeats the cycle every few hundred milliseconds.
The company says this approach helps robots handle situations that often cause failures in existing systems. Many industrial robots still rely on fixed, pre-programmed paths and operate best in structured environments. Even newer AI systems based on vision-language-action models can struggle when objects move, layouts change, or workflows vary.
Rhoda trains its models first on large collections of video data and later refines them with robot-specific learning. The initial training uses hundreds of millions of online videos so the system can learn motion patterns, physics, and physical interactions before controlling a robot.
After that stage, the model is fine-tuned with smaller sets of real robot data. This step teaches the system how to convert visual predictions into physical actions.
Rhoda refers to this architecture as a Direct Video Action model. Instead of creating a plan once and executing it without feedback, the system keeps updating actions as it observes new information from the environment.
According to the company, this closed-loop process allows robots to adjust to changes while maintaining accuracy. It also reduces the amount of robot training data required. Rhoda says some new tasks can be learned using about ten hours of teleoperation data.
The technology has already been tested in manufacturing settings where robots must handle changing materials and workflows. In one high-volume production evaluation, a robot using the system completed a component-processing cycle in under two minutes without human intervention.
Rhoda plans to develop FutureVision as a foundation model that can be licensed to companies building robotic hardware and software platforms.
“We believe the next era of robotics requires models that understand how the world moves – not just what it looks like or how it’s described in language,” said Jagdeep Singh, cofounder and CEO of Rhoda.
